Claude Code voice: three ways to talk to Claude Code
Dictate a prompt, hold a spoken conversation, or speak from any app while you point at the problem.
Summary
There are three ways to talk to Claude Code today. The built-in /voice command turns speech into prompt text. VoiceMode, a community MCP server, adds a spoken back-and-forth inside the session. Denker takes a spoken request from any app, together with what you point at on screen, and reads the reply aloud. Pick by what you want to say it to and whether you want to hear an answer.
Key points
- Only want to stop typing prompts in the terminal: use /voice. It comes with your Claude.ai sign-in.
- /voice does not work over SSH, in cloud sessions, or with an API key, Amazon Bedrock, Google Cloud or Microsoft Foundry.
- Want a spoken conversation inside a Claude Code session: VoiceMode, an open-source MCP server.
- Want to speak from any app, point at what you mean and hear the reply: Denker, on Mac and Windows.
Yes, Claude Code has built-in voice: run /voice, hold Space, and your speech is typed into the prompt as you talk. It is dictation only. Anthropic's docs describe speech to text for your prompt, not spoken replies, and it needs a Claude.ai sign-in and a local microphone. If you want Claude Code to answer out loud, or to talk to it from outside the terminal, you need something on top: a community MCP server such as VoiceMode, or a desktop app such as Denker. This guide is written by Denker and based on each tool's own pages, checked November 20, 2026. We have not benchmarked the tools.
How do the ways to talk to Claude Code compare?
The table compares what each option documents, not measured accuracy or speed. Operating-system dictation is included because it is already on your computer and types into any text field.
| Option | What you talk to | Talks back | Works outside the terminal | Point at the screen | Setup | Price |
|---|---|---|---|---|---|---|
| Built-in /voice | The Claude Code prompt, in the CLI and the VS Code extension | No. Speech to text only | No | No | Run /voice. Needs a Claude.ai sign-in and a local microphone | Comes with a Claude.ai sign-in. Transcription does not count toward usage limits |
| System dictation (macOS Dictation, Windows voice typing) | Any place you can type text | No | Yes | No | Mac: turn on Dictation in Keyboard settings. Windows: press Win+H | Part of macOS and Windows |
| VoiceMode (community) | Claude Code and other MCP-capable agents | Yes | No. It is an MCP server inside the agent session | Not described in its README | Install the Claude Code plugin or the Python package. Needs an OpenAI API key or local Whisper and Kokoro | Open source (MIT). Cloud speech is billed to your OpenAI key |
| Denker | Denker, which runs on your own Claude Code, Codex or OpenCode | Yes. Replies read aloud, and a live back-and-forth | Yes. A shortcut starts a request from any app | Yes. Draw marks on the screen while you talk | Install the app on Mac or Windows and connect your AI tool | Free: 25 on-screen requests a month, dictation anywhere unlimited. Pro: €19 a month |
How do you turn on voice in Claude Code?
Open Claude Code in a terminal on the computer your microphone is plugged into. Run /voice. Claude Code checks the microphone, and on macOS your terminal asks for microphone permission the first time. Then hold Space and speak. Your words appear in the prompt, dimmed until the transcript is final. Release Space and press Enter to send.
There are two modes. Hold mode is the default: recording runs while you hold the key. Tap mode starts on one tap and sends the prompt on the next, as long as the transcript has at least three words. In tap mode a recording also stops after 15 seconds of silence or two minutes in total. Press Esc or Ctrl+C to cancel a recording.
The setting persists across sessions. You can also set it in your user settings file with a voice object, and add autoSubmit to send a hold-mode prompt when you release the key. To move dictation off Space, rebind voice:pushToTalk in ~/.claude/keybindings.json. A modifier combination such as meta+k starts recording on the first keypress, with no warmup.
/voice toggle dictation on or off
/voice hold hold the key while you speak (default)
/voice tap tap to start, tap again to send
/voice off turn dictation offWhat does /voice need, and where does it not work?
Anthropic's docs list three requirements. You must be signed in with a Claude.ai account. You need a local microphone. And if you run Claude Code in WSL, you need WSLg, which comes with WSL2 from the Microsoft Store.
That rules out some setups. /voice is not available when Claude Code uses an Anthropic API key directly, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. It does not work in cloud sessions or over SSH. In the VS Code extension it does not work in Remote sessions such as SSH, Dev Containers and Codespaces. An organization can also turn it off by policy.
Your audio is streamed to Anthropic's servers for transcription. It is not processed on your computer. Transcription does not use Claude messages or tokens and does not count toward the limits shown in /usage.
Recording works on macOS, Linux and Windows. On Linux, Claude Code falls back to arecord or SoX if its own audio module cannot load.
- Signed in with a Claude.ai account, not an API key or a cloud provider
- Claude Code runs on the computer with the microphone, not over SSH
- Your terminal has microphone permission
- The language setting matches the language you speak
Which languages does Claude Code voice understand?
The docs list 20 dictation languages: Czech, Danish, Dutch, English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Turkish and Ukrainian.
Dictation follows the same language setting that controls Claude's reply language. Set it in /config. If it is empty, dictation uses English. If your language is not on the list, /voice warns you and falls back to English for dictation.
Can Claude Code talk back?
Not with /voice. Anthropic's voice dictation page covers one direction: your speech becomes text in the prompt. Claude Code answers in text.
VoiceMode fills that gap from the community side. Its README describes it as natural voice conversations with Claude Code and other MCP-capable agents: you speak, you hear the response, and recording stops when you stop speaking. It lists Linux, macOS, Windows (native or WSL) and NixOS, and needs Python 3.10 to 3.14 and a computer with a microphone and speakers. Speech runs through an OpenAI API key or through local Whisper and Kokoro. The recommended install is a Claude Code plugin, shown below. It is a community project, so check its README for current notes before you install.
claude plugin marketplace add mbailey/voicemode
claude plugin install voicemode@voicemode
/voicemode:install
/voicemode:converseWhat works well by voice, and what does not?
This section is our view from building a voice product, not a test result. Voice is good for intent. It is poor for exact characters.
It works when you describe a bug you can see: "the menu closes before I can click the second item, find out why". It works for review: "walk me through what changed in the auth middleware and what could break". It works for small changes you can name: "rename this prop to isOpen everywhere and run the tests". In each case you say what you want and the agent finds the code.
It does not work for dictating code. Brackets, operators, casing and file paths are slow to say and easy to get wrong. Type those, or let the agent write them. Anthropic's docs say Claude Code's transcription is tuned for coding vocabulary such as regex, OAuth, JSON and localhost, and that it adds your project and git branch names as hints. That helps with terms. It does not make speaking a line of code faster than typing it.
A habit that helps: speak the request, then type the one exact thing. /voice inserts the transcript at the cursor, so you can mix both in one prompt.
Where does Denker fit?
Denker is for the moment the problem is not in the terminal. The broken layout is in the browser. The wrong number is in a spreadsheet. With /voice you would switch to the terminal and describe in words what you are looking at. With Denker you stay where you are: hold Ctrl+Option on a Mac or Ctrl+Win on Windows, say what you want, and Shift-drag to draw a mark around the thing you mean. Denker sends your words and the marked screenshot to the agent you use, Claude Code, Codex or OpenCode, on your own account.
It also answers. Turn on Read replies aloud in Settings and Denker speaks its answer. Esc stops it. Hold the voice shortcut for 2.5 seconds and you get a live conversation with no key press per turn. Talk over a reply and it stops, and what you said becomes the next request.
Dictation is separate and has no limit on the Free plan. Double-tap Cmd on a Mac, or press Win+Shift on Windows, and Denker types what you say into any app, your terminal included.
Two limits to know. Screenshots need a model that can read images. And a voice request goes to your Denker chat, not into a terminal session you already have open.
When is /voice enough?
If you only want to speak prompts into the terminal instead of typing them, use /voice and stop there. It is built in, it comes with your Claude.ai sign-in, and it takes one command to turn on. Denker adds nothing you need for that.
If you work over SSH or sign in with an API key, /voice is not available. System dictation or a dictation app that types into the terminal is the simplest replacement. If you want to hear answers inside the session, try VoiceMode. If your requests start from what is on your screen, in any app, that is the case Denker is built for.
Frequently asked questions
Does Claude Code have a voice mode?
Yes. Claude Code has built-in voice dictation. Run /voice, then hold Space while you speak, or use tap mode. Your speech is transcribed into the prompt. It is speech to text; the docs do not describe spoken replies.
Can Claude Code talk back?
Not with the built-in /voice command, which only turns your speech into prompt text. For spoken replies you need an add-on. VoiceMode is a community MCP server for voice conversations with Claude Code. Denker reads replies aloud and has a live voice conversation, running on your own Claude Code.
Does Claude Code voice work on Windows?
Yes. Anthropic's docs say audio recording works on macOS, Linux and Windows. In WSL you need WSLg, which is included with WSL2 from the Microsoft Store. Without it, run Claude Code in native Windows.
Does /voice work over SSH or in remote sessions?
No. Voice dictation needs a local microphone. Anthropic's docs say it does not work in cloud sessions or SSH sessions, and the VS Code extension does not support it in Remote sessions such as SSH, Dev Containers and Codespaces.
Is Claude Code voice free?
It has no separate charge in Anthropic's docs. It needs a Claude.ai account sign-in, and transcription does not use Claude messages or tokens or count toward the limits shown in /usage. It is not available with an Anthropic API key, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry.
Which languages does Claude Code voice support?
Anthropic's docs list 20: Czech, Danish, Dutch, English, French, German, Greek, Hindi, Indonesian, Italian, Japanese, Korean, Norwegian, Polish, Portuguese, Russian, Spanish, Swedish, Turkish and Ukrainian. Set the language in /config. The default is English.
Is my audio processed on my computer?
Not with /voice. Anthropic's docs say recorded audio is streamed to Anthropic's servers for transcription and is not processed locally. VoiceMode can run local Whisper and Kokoro, and Denker's dictation has local models that run on the device.
Were these tools benchmarked?
No. This Denker-authored guide compares each tool's own pages, checked November 20, 2026. The section on what works by voice is our opinion, not a measurement.